Changes in version 0.99.0 (2025-10-21)                 

Overview

CrcBiomeScreen is an R package designed to streamline microbiome-based
colorectal cancer (CRC) screening workflows.
It provides standardized functions for preprocessing, taxonomic data
handling, machine learning model training,
and cross-cohort validation — supporting reproducible and interpretable
microbiome analysis for biomarker discovery.

This version marks the first public release of the package, submitted to
Bioconductor.

New Features

Data Processing

  - SplitTaxas() now automatically detects taxonomy separators
    (supports |, ., _, and ;) to handle common formats from MetaPhlAn,
    QIIME, etc.
    Adds:
    
      - OriginalTaxa column to retain the raw taxonomy string.
      - Special handling for uncultured and unclassified taxa
        (e.g. converts f__Rikenellaceae|g__unclassified →
        Rikenellaceae_unclassified).

  - KeepTaxonomicLevel() filters data at a user-defined rank (e.g.,
    Genus, Family, Species)
    and automatically collapses lower-level abundances.
    Handles multiple nested unclassified levels (e.g.,
    D_2__Clostridia.D_3__uncultured.D_4__uncultured).

Machine Learning

  - TrainModel() serves as a unified interface for training Random
    Forest (RF) and XGBoost classifiers.
    
      - Integrates with internal preprocessing and class-weight
        balancing.
      - Uses withr::with_seed() for local reproducibility instead of
        set.seed().
      - Adds support for both training and external validation
        workflows.

  - EvaluateRF() and EvaluateXGBoost() now return standardized
    performance metrics
    (AUC, accuracy, recall, F1) and store model outputs in the main
    object structure.

  - ValidateModelOnData() supports model evaluation across independent
    datasets.

Data Object Structure

  - Introduced the CrcBiomeScreenObject class to store:
      - Absolute and relative abundance tables
      - Taxonomic annotations
      - Model training results
      - Validation data and predictions
      - Visual outputs (e.g., ROC, variable importance)

This ensures data provenance and reproducibility across the full
workflow.

Utility and Visualization

  - Added built-in plotting functions for:
      - Model ROC curves (PlotAUC)

Documentation and Testing

  - Added comprehensive vignette:
    
      - Demonstrates preprocessing, genus-level filtering, model
        training, and validation.
      - Includes examples for multiple taxonomy formats and classifier
        comparisons.

  - Implemented unit tests under tests/testthat/ for key components:
    
      - Taxa splitting
      - Level filtering
      - Model reproducibility

Technical and Compliance Updates

  - Adopted MIT license.
  - Added BugReports and URL fields in DESCRIPTION.
  - Removed unnecessary system files (.Rproj, .DS_Store).
  - Replaced all instances of set.seed() with withr::with_seed() for
    Bioconductor compliance.
  - Reduced hard-coded dependencies and improved optional imports.

Future Plans

  - Add support for additional classifiers.
  - Integrate feature interpretation.
  - Provide reproducible benchmarking across public CRC datasets.

Version 0.99.13

  - Fixed vignette and example issues
  - Temporarily disabled XGBoost execution due to compatibility issues
    with caret
  - Improved documentation and S4 accessor usage

Version 0.99.14(2026-04-10)

  - runnable toy workflow added to vignette
  - examples updated for TrainModels, ValidateModelOnData, RunScreening
  - documentation for included datasets completed

Version 0.99.15(2026-04-14)

  - Standardized the format of the vignette and documentation

Maintainer: Chengxin Li (University of Leeds)
Date: 2025-10-21
License: MIT
Repository: https://github.com/omicsForestry/CrcBiomeScreen